Improving MSVM-RFE for Multiclass Gene Selection∗

Abstract Along with the advent of DNA microarray technology, gene expression profiling has been widely used to study molecular signatures of many diseases and to develop molecular diag-nostics for disease prediction. In class prediction problems using expression data, gene selection is essential to improve the prediction accuracy and to identify informative genes for a disease. In this paper we improve the multi-class support vector machine-recursive feature elimination (MSVM-RFE) by combining minimum redundancy maximum relevancy (mRMR) criterion and introducing the kernel. The result is the better performance with a smaller number of irredundant genes for multi-class datasets.